Stage division and fault detection method based on correlation analysis

Abstract

The invention relates to a stage division and fault detection method based on correlation analysis. The method uses the change degree of information in the time slice matrix of a stable stage and a transition stage of an industrial process to identify multiple stages of the reaction duration of the whole industrial process. The method comprises the steps of arranging and standardizing historical batch data in a variable expansion mode, unfolding according to a batch mode, the unfolded time slice matrix comprising time-varying characteristics of an industrial process, and carrying out time sequence stage division according to distribution characteristics of evaluation values. After phase division is carried out, a model is established for each duration stage to monitor quality-related faults, carry out residual information subspace extraction on the industrial process variables, and monitor the abnormity irrelevant to quality in the industrial process. The method is applied to an industrial penicillin fermentation industrial process, which shows that the method has better monitoring performance and forecast capability.

Description

technical field [0001] The invention belongs to the technical field of automatic control, and relates to a phase division and fault detection method based on correlation analysis. Background technique [0002] The performance monitoring of industrial processes generally starts from historical production data, establishes corresponding performance monitoring and fault detection models through statistical information processing methods, and uses this model to monitor the industrial process of product production, discover and eliminate abnormalities in industrial processes in time conditions, so that the industrial process of production can run efficiently, safely and stably. Traditional analytical model-based industrial process monitoring requires accurate mathematics and production experience, which limits its practical application. Multi-way Principle Component Analysis (MPCA) and multi-way partial least squares (Multi-way partial Least Squares, MPLS) are commonly used stat...

AI Technical Summary

Problems solved by technology

The basic idea of ​​the cluster-based stage identification method is that the change of the industrial process relationship is related to the switching of multiple stages. This type of method can effectively capture the characteristics of the dynamic industrial process without filling unknown and missing data, but this type of method still exists. Some deficiencies, for example, the multi-stage division of the clustering algorithm cannot consider the order and timing of the stages, and it is prone to time series crossover.

In addition, the stage division algorithm ignores the stage transfer characteristics of the transition stage, and strictly divides each sample into a specific stage, which will easily lead to a decrease in the accuracy of the stage representative model, and the cluster-based stage division method is often limited by the parameters Choices such as initial center, initial number of clusters, minimum stage duration length

Images